Data processing method, device, electronic device and readable storage medium

Abstract

The present disclosure discloses a data processing method, apparatus, electronic device, and readable storage medium. The data processing method includes: obtaining object feature data of an object, user portrait data of a user, and conversation text data of the user; obtaining an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data; and determining the probability of the user performing an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a data processing method, device, electronic device, and readable storage medium. Background Art

[0002] When recommending products or services, neural networks can be used to predict the probability of a user purchasing a product or service based on its characteristics and user profile, thereby improving the accuracy and efficiency of recommendations. However, this prediction method relies primarily on static and historical data and may not accurately reflect users' current needs and preferences. Summary of the Invention

[0003] In order to solve the problems in the related art, the embodiments of the present disclosure provide a data processing method, device, electronic device and readable storage medium.

[0004] In a first aspect, an embodiment of the present disclosure provides a data processing method.

[0005] Specifically, the data processing method includes:

[0006] Obtaining object feature data of an object, user portrait data of a user, and conversation text data of the user;

[0007] Obtaining an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data;

[0008] Based on the object feature vector, the user portrait vector, and the conversation text vector, a probability of the user performing an action related to the object is determined.

[0009] In combination with the first aspect, in a first implementation of the first aspect of the present disclosure, obtaining the object feature vector corresponding to the object feature data includes:

[0010] If the object feature data belongs to category data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data;

[0011] If the object feature data is numerical data, an object feature vector corresponding to the object feature data is determined according to the value of the object feature data and a first preset random vector.

[0012] The object feature vectors corresponding to different object feature data are different.

[0013] In combination with the first aspect, in a second implementation of the first aspect of the present disclosure, obtaining a user portrait vector corresponding to the user portrait data includes:

[0014] If the user portrait data belongs to category data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data;

[0015] If the user portrait data is numerical data, a user portrait vector corresponding to the user portrait data is determined according to the value of the user portrait data and a second preset random vector.

[0016] Among them, the user portrait vectors corresponding to different user portrait data are different.

[0017] In combination with the first aspect, in a third implementation of the first aspect of the present disclosure, obtaining a conversation text vector corresponding to the conversation text data includes:

[0018] Obtaining a current text vector corresponding to a current dialogue text in the dialogue text data;

[0019] A previous text vector corresponding to a previous dialogue text preceding the current dialogue text in the dialogue text data is obtained.

[0020] In conjunction with the first aspect, in a fourth implementation of the first aspect of the present disclosure, determining the probability that the user performs an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector includes:

[0021] Obtaining, through the first model, a linear combination result of a first projection vector of the object feature vector in the latent space of the conversation text vector, a second projection vector of the user portrait vector in the latent space, and the conversation text vector;

[0022] Obtaining a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector through a second model;

[0023] Obtaining semantic information representation of the conversation text vector through a third model;

[0024] The probability that the sample user performs a behavior related to the object is determined through a fourth model according to the linear combination result, the nonlinear combination result, and the semantic information representation.

[0025] Combined with the fourth implementation manner of the first aspect, in the fifth implementation manner of the first aspect of the present disclosure, the obtaining of the linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the first model includes:

[0026] Through the first model, assign randomly initialized and trainable weight coefficients to each of the first projection vector, the second projection vector, and the dialogue text vector, and output the weighted linear combination result of each vector in the first projection vector, the second projection vector, and the dialogue text vector as the linear combination result.

[0027] Combined with the fourth implementation manner of the first aspect, in the sixth implementation manner of the first aspect of the present disclosure, the first projection vector, the second projection vector, and the dialogue text vector include N vectors. The obtaining of the non - linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the second model includes:

[0028] For k = 0, 1, …… K, where K>1: perform an outer product operation on each vector in the k - th order initial matrix with the N vectors respectively to obtain N k - th order feature matrices; obtain the k - th order result matrix from the N k - th order feature matrices, and obtain the k - th order result vector based on the k - th order result matrix; when k < K, use the k - th order result matrix as the (k + 1) - th order initial matrix, where the 0 - th order initial matrix includes the N vectors;

[0029] Based on the 0 - th order result vector to the K - th order result vector, obtain the non - linear combination result.

[0030] Combined with the sixth implementation manner of the first aspect, in the seventh implementation manner of the first aspect of the present disclosure, the obtaining of the k - th order result matrix from the N k - th order feature matrices includes obtaining the k - th order result matrix from the N k - th order feature matrices through a convolutional neural network or a fully - connected network;

[0031] The obtaining of the k - th order result vector based on the k - th order result matrix includes performing pooling processing on the k - th order result matrix to obtain the k - th order result vector.

[0032] Combined with the sixth implementation manner of the first aspect, in the eighth implementation manner of the first aspect of the present disclosure, the obtaining of the non - linear combination result based on the 0 - th order result vector to the K - th order result vector includes:

[0033] Concatenate the 0 - th order to the K - th order result vectors to obtain a concatenated vector;​​​

[0035] In conjunction with the fourth implementation of the first aspect, in a ninth implementation of the first aspect of the present disclosure, obtaining the semantic information representation of the conversation text vector using the third model includes:

[0036] The conversation text vector is input into a second fully connected network to obtain a semantic information representation of the conversation text vector.

[0037] In conjunction with the fourth implementation manner of the first aspect, in a tenth implementation manner of the first aspect of the present disclosure, determining, by the fourth model, based on the linear combination result, the nonlinear combination result, and the semantic information representation, the probability that the user performs an action related to the object includes:

[0038] The linear combination result, the nonlinear combination result, and the semantic information representation are input into a third fully connected network to obtain a probability that the user performs an action related to the object.

[0039] In combination with the fourth implementation manner of the first aspect, in an eleventh implementation manner of the first aspect of the present disclosure, the method further includes:

[0040] Acquire sample data, the sample data including object feature data of a sample object, user portrait data of a sample user, conversation text data of the sample user, and a record of whether the user performs any behavior related to the object;

[0041] The first model, the second model, the third model, and the fourth model are trained using the sample data.

[0042] In combination with the first aspect, in a twelfth implementation manner of the first aspect of the present disclosure, the object is a commodity, and the behavior is purchasing or clicking on the commodity; or the object is a service, and the behavior is purchasing the service.

[0043] In a second aspect, an embodiment of the present disclosure provides a data processing device.

[0044] Specifically, the data processing device includes:

[0045] A first acquisition module is configured to acquire object feature data of an object, user portrait data of a user, and conversation text data of the user;

[0046] A second acquisition module is configured to acquire an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data;

[0047] The determination module is configured to determine the probability that the user performs an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector.

[0048] In conjunction with the second aspect, in a first implementation of the second aspect of the present disclosure, obtaining the object feature vector corresponding to the object feature data includes:

[0049] If the object feature data belongs to category data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data;

[0050] If the object feature data is numerical data, an object feature vector corresponding to the object feature data is determined according to the value of the object feature data and a first preset random vector.

[0051] The object feature vectors corresponding to different object feature data are different.

[0052] In conjunction with the second aspect, in a second implementation of the second aspect of the present disclosure, obtaining a user portrait vector corresponding to the user portrait data includes:

[0053] If the user portrait data belongs to category data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data;

[0054] If the user portrait data is numerical data, a user portrait vector corresponding to the user portrait data is determined according to the value of the user portrait data and a second preset random vector.

[0055] Among them, the user portrait vectors corresponding to different user portrait data are different.

[0056] In conjunction with the second aspect, in a third implementation of the second aspect of the present disclosure, obtaining a conversation text vector corresponding to the conversation text data includes:

[0057] Obtaining a current text vector corresponding to a current dialogue text in the dialogue text data;

[0058] A previous text vector corresponding to a previous dialogue text preceding the current dialogue text in the dialogue text data is obtained.

[0059] In combination with the second aspect, in the fourth implementation manner of the second aspect of the present disclosure, determining the probability that the user performs an action related to the object based on the object feature vector, the user profile vector, and the conversation text vector includes:

[0060] Through a first model, obtain a first projection vector of the object feature vector in the latent space of the conversation text vector, a second projection vector of the user profile vector in the latent space, and a linear combination result of the conversation text vector;

[0061] Through a second model, obtain a non-linear combination result of the first projection vector, the second projection vector, and the conversation text vector;

[0062] Through a third model, obtain a semantic information representation of the conversation text vector;

[0063] Through a fourth model, determine the probability that the sample user performs an action related to the object according to the linear combination result, the non-linear combination result, and the semantic information representation.

[0064] In combination with the fourth implementation manner of the second aspect, in the fifth implementation manner of the second aspect of the present disclosure, the step of obtaining, through the first model, a linear combination result of the first projection vector, the second projection vector, and the conversation text vector includes:

[0065] Through the first model, assign randomly initialized and trainable weight coefficients to each of the first projection vector, the second projection vector, and the conversation text vector, and output a weighted linear combination result of each vector in the first projection vector, the second projection vector, and the conversation text vector as the linear combination result.

[0066] In combination with the fourth implementation manner of the second aspect, in the sixth implementation manner of the second aspect of the present disclosure, the first projection vector, the second projection vector, and the conversation text vector include N vectors. The step of obtaining, through the second model, a non-linear combination result of the first projection vector, the second projection vector, and the conversation text vector includes:

[0067] For k = 0, 1,..., K, where K > 1: perform an outer product operation on each vector in the k-th order initial matrix with the N vectors respectively to obtain N k-th order feature matrices; obtain a k-th order result matrix from the N k-th order feature matrices, and obtain a k-th order result vector based on the k-th order result matrix; when k < K, use the k-th order result matrix as the (k + 1)-th order initial matrix, where the 0-th order initial matrix includes the N vectors;

[0068] Based on the 0th order result vector to the Kth order result vector, the nonlinear combination result is obtained.

[0069] In combination with the sixth implementation manner of the second aspect, in a seventh implementation manner of the second aspect of the present disclosure, obtaining the k-th order result matrix from the N k-th order feature matrices includes obtaining the k-th order result matrix from the N k-th order feature matrices through a convolutional neural network or a fully connected network;

[0070] The obtaining of the k-th order result vector based on the k-th order result matrix includes performing pooling processing on the k-th order result matrix to obtain the k-th order result vector.

[0071] In combination with the sixth implementation manner of the second aspect, in an eighth implementation manner of the second aspect of the present disclosure, obtaining the nonlinear combination result based on the 0th-order result vector to the Kth-order result vector includes:

[0072] Splicing the 0th to Kth order result vectors to obtain a spliced vector;

[0073] The concatenated vector is input into a first fully connected network to obtain the nonlinear combination result.

[0074] In conjunction with the fourth implementation of the second aspect, in a ninth implementation of the second aspect of the present disclosure, obtaining the semantic information representation of the conversation text vector using the third model includes:

[0075] The conversation text vector is input into a second fully connected network to obtain a semantic information representation of the conversation text vector.

[0076] In conjunction with the fourth implementation manner of the second aspect, in a tenth implementation manner of the second aspect of the present disclosure, determining, by the fourth model, based on the linear combination result, the nonlinear combination result, and the semantic information representation, the probability that the user performs an action related to the object includes:

[0077] The linear combination result, the nonlinear combination result, and the semantic information representation are input into a third fully connected network to obtain a probability that the user performs an action related to the object.

[0078] In combination with the fourth implementation manner of the second aspect, in an eleventh implementation manner of the second aspect of the present disclosure, the apparatus further includes:

[0079] a third acquisition module configured to acquire sample data, the sample data including object feature data of a sample object, user portrait data of a sample user, conversation text data of the sample user, and a record of whether the user performs any behavior related to the object;

[0080] A training module is configured to train the first model, the second model, the third model, and the fourth model using the sample data.

[0081] In combination with the second aspect, in an eleventh implementation manner of the second aspect of the present disclosure, the object is a commodity, and the behavior is purchasing or clicking on the commodity; or the object is a service, and the behavior is purchasing the service.

[0082] In a third aspect, an embodiment of the present disclosure provides an electronic device comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, and wherein the one or more computer instructions are executed by the processor to implement a method as described in any one of the first aspect to the twelfth implementation of the first aspect.

[0083] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the methods described in the first aspect to the twelfth implementation of the first aspect are implemented.

[0084] According to the technical solution provided by the embodiments of the present disclosure, object feature data of an object, user portrait data of a user, and conversation text data of the user are obtained, an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data are obtained, and based on the object feature vector, the user portrait vector, and the conversation text vector, the probability of the user performing an action related to the object is determined.

[0085] According to the embodiments of the present disclosure, based on the object feature data of the object, the user's user portrait data and the user's conversation text data, the probability of the user performing actions related to the object is predicted, which can fully utilize the user's current needs and preferences reflected in the conversation text data and improve the real-time and accuracy of the prediction.

[0086] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:

[0088] Figure 1 A flow chart showing a data processing method according to an embodiment of the present disclosure is shown.

[0089] Figure 2 A system architecture diagram of a data processing method according to an embodiment of the present disclosure is shown.

[0090] Figure 3 A structural block diagram of a data processing device according to an embodiment of the present disclosure is shown.

[0091] Figure 4 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0092] Figure 5 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0093] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0094] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, actions, components, parts, or combinations thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, actions, components, parts, or combinations thereof exist or are added.

[0095] It should also be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0096] In this disclosure, the acquisition of user information or user data is an operation authorized and confirmed by the user, or actively selected by the user.

[0097] As mentioned above, when recommending products or services, neural networks can be used to predict the probability of a user purchasing a product or service based on its characteristics and user profile, thereby improving the accuracy and efficiency of recommendations. However, this prediction method primarily relies on static and historical data and may not accurately reflect users' current needs and preferences.

[0098] In recent years, the application of intelligent conversational systems based on deep learning has become increasingly widespread. Intelligent conversational robots and personal assistants have played a significant role in areas such as home use and corporate customer service. Users may express clearer current needs and preferences during conversations. Therefore, leveraging the conversational process can help quickly and accurately identify user needs and clarify user intent and purchase intentions. The disclosed embodiments, by incorporating conversational text information into user behavior predictions, can effectively improve the real-time and accuracy of predictions.

[0099] An embodiment of the present disclosure provides a data processing method, comprising: obtaining object feature data of an object, user portrait data of a user, and conversation text data of the user; obtaining an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data; and determining, based on the object feature vector, the user portrait vector, and the conversation text vector, the probability that the user performs an action related to the object.

[0100] According to the embodiments of the present disclosure, based on the object feature data of the object, the user's user portrait data and the user's conversation text data, the probability of the user performing actions related to the object is predicted, which can fully utilize the user's current needs and preferences reflected in the conversation text data and improve the real-time and accuracy of the prediction.

[0101] The solution according to the embodiments of the present disclosure can be used to improve the accuracy of prediction models such as click-through rate and purchase rate in marketing recommendation and shopping guide scenarios, which is of great significance to the intelligence of marketing recommendation and shopping guide systems. It can also be used to predict the probability of users watching recommended videos or articles, predict the probability of users booking recommended hotels or restaurants, or predict the user's satisfaction with the processing results and the possible actions they may take next when handling user complaints, etc.

[0102] Figure 1 1 is a flow chart showing a data processing method according to an embodiment of the present disclosure. Figure 1 As shown, the data processing method includes the following steps S101-S103:

[0103] In step S101, object feature data of an object, user portrait data of a user, and conversation text data of the user are obtained;

[0104] In step S102, an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data are obtained;

[0105] In step S103, based on the object feature vector, the user portrait vector, and the conversation text vector, the probability of the user performing an action related to the object is determined.

[0106] According to the embodiments of the present disclosure, based on the object feature data of the object, the user's user portrait data and the user's conversation text data, the probability of the user performing actions related to the object is predicted, which can fully utilize the user's current needs and preferences reflected in the conversation text data and improve the real-time and accuracy of the prediction.

[0107] According to an embodiment of the present disclosure, the object is a commodity, and the behavior is purchasing or clicking the commodity; or the object is a service, and the behavior is purchasing the service.

[0108] According to an embodiment of the present disclosure, a user's conversation text data may be the conversation text data between the user and a human or machine customer service representative, and may include data entered in text form or text data converted from voice conversation data. According to an embodiment of the present disclosure, the user's conversation text data may include identification information that identifies the speaker of a text segment. For example, the identification information can be used to determine whether a text segment is a user's speech or a customer service representative's speech. This identification information can more accurately reflect the user's intent and avoid errors caused by confusing the speaker.

[0109] According to the embodiments of the present disclosure, the probability of a user purchasing or clicking on a product or a user purchasing a service can be predicted in combination with the user's conversation text data, thereby improving the real-time performance and accuracy of the prediction.

[0110] According to embodiments of the present disclosure, object feature data may include, for example, one or more of the following data about the object: name, identifier, price, rating, sales volume, provider, etc., but is not limited thereto. According to embodiments of the present disclosure, user profile data may include, for example, one or more of the following data about the user: name, gender, age, region, purchase history, click history, etc., but is not limited thereto.

[0111] Reference below Figure 2 The method according to the embodiment of the present disclosure is described in detail.

[0112] Figure 2 A system architecture diagram of a data processing method according to an embodiment of the present disclosure is shown.

[0113] According to an embodiment of the present disclosure, obtaining the object feature vector corresponding to the object feature data includes: when the object feature data belongs to category feature data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data; when the object feature data belongs to numerical feature data, determining the object feature vector corresponding to the object feature data according to the value of the object feature data and a first preset random vector, wherein the object feature vectors corresponding to different object feature data are different.

[0114] According to an embodiment of the present disclosure, obtaining a user portrait vector corresponding to the user portrait data includes: when the user portrait data belongs to category feature data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data; when the user portrait data belongs to numerical feature data, determining the user portrait vector corresponding to the user portrait data according to the value of the user portrait data and a second preset random vector, wherein the user portrait vectors corresponding to different user portrait data are different.

[0115] like Figure 2 As shown, the object feature data "place of origin" belongs to categorical feature data. The value of "place of origin" can be, for example, "Beijing," "Shanghai," etc. Based on its value, it can be mapped to a corresponding random vector, which is used as the object feature vector. Different places of origin correspond to different object feature vectors. The object feature data "price" belongs to numerical feature data. Its value can be, for example, 30 yuan, 50 yuan, etc. For numerical object feature data, its value can be multiplied by a first preset random vector to obtain the corresponding object feature vector. Different prices correspond to different object feature vectors.

[0116] like Figure 2 As shown, the user portrait data "credit rating" belongs to category feature data. The value of "credit rating" can be, for example, "level one", "level two", etc. According to its value, it can be mapped to the corresponding random vector, and the corresponding random vector is used as the object feature vector. Different credit ratings correspond to different object feature vectors. Figure 2 Although not shown, user profile data may also include numerical feature data, such as "age," which may have values such as 20, 30, and so on. For numerical user profile data, its value may be multiplied by a second preset random vector to obtain a corresponding user profile vector, with different ages corresponding to different user profile vectors.

[0117] According to an embodiment of the present disclosure, the obtaining of the conversation text vector corresponding to the conversation text data includes: obtaining a current text vector corresponding to the current conversation text in the conversation text data; and obtaining a previous text vector corresponding to the conversation text before the current conversation text in the conversation text data.

[0118] According to embodiments of the present disclosure, the current conversation text may be a single sentence currently input by the user, or multiple sentences currently input continuously. The previous conversation text may be one or more sentences previously input by the user and / or customer service, and the conversation text vector includes the current text vector and the previous text vector.

[0119] like Figure 2 As shown, according to an embodiment of the present disclosure, the previous conversation text can be segmented into characters or words to obtain corresponding character vectors or word vectors, and then these character vectors or word vectors are encoded by, for example, a convolutional neural network (CNN) or a recurrent neural network (RNN) to obtain the previous text vector. According to an embodiment of the present disclosure, the current conversation text can be segmented into characters or words to obtain corresponding character vectors or word vectors, and then these character vectors or word vectors are encoded by, for example, a convolutional neural network (CNN) or a recurrent neural network (RNN) to obtain the current text vector.

[0120] According to an embodiment of the present disclosure, the method of determining the probability that the user performs a behavior related to the object based on the object feature vector, the user portrait vector, and the conversation text vector includes: obtaining, through a first model, a linear combination result of a first projection vector of the object feature vector in the latent space of the conversation text vector, a second projection vector of the user portrait vector in the latent space, and the conversation text vector; obtaining, through a second model, a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector; obtaining, through a third model, a semantic information representation of the conversation text vector; and determining, through a fourth model, the probability that the user performs a behavior related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation.

[0121] like Figure 2 As shown, the object feature vector is projected into the latent space of the conversation text vector to obtain a first projection vector, and the user profile vector is projected into the latent space of the conversation text vector to obtain a second projection vector. A first model is used to obtain a linear combination of the first projection vector, the second projection vector, and the conversation text vector. A second model is used to obtain a nonlinear combination of the first projection vector, the second projection vector, and the conversation text vector. A third model is used to obtain a semantic information representation of the conversation text vector. A fourth model is used to determine the probability of the user performing an action related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation.

[0122] According to the embodiments of the present disclosure, the object features, user portrait features and explicit patterns of the conversation text are expressed through a linear combination of the first projection vector, the second projection vector and the conversation text vector, the interactive information of the object features, user portrait features and the conversation text is mined through a nonlinear combination of the first projection vector, the second projection vector and the conversation text vector, and the prediction results are enhanced through the semantic information representation of the conversation text vector. This can fully utilize the explicit, implicit and interactive information of the input data and improve the real-time and accuracy of the prediction results.

[0123] As shown in Figure 2 FIG. [ID], according to an embodiment of the present disclosure, obtaining the linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the first model includes: through the first model, assigning randomly initialized and trainable weight coefficients to each of the first projection vector, the second projection vector, and the dialogue text vector, and outputting the weighted linear combination result of each vector in the first projection vector, the second projection vector, and the dialogue text vector as the linear combination result.

[0124] According to an embodiment of the present disclosure, the first projection vector, the second projection vector, and the dialogue text vector include N vectors. Obtaining the non - linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the second model includes:

[0125] For k = 0, 1, …… K, where K>1: performing an outer product operation on each vector in the k - th order initial matrix with the N vectors respectively to obtain N k - th order feature matrices; obtaining a k - th order result matrix from the N k - th order feature matrices, and obtaining a k - th order result vector based on the k - th order result matrix; when k<K, taking the k - th order result matrix as the (k + 1) - th order initial matrix, where the 0 - th order initial matrix includes the N vectors;

[0126] Obtaining the non - linear combination result based on the 0 - th order result vector to the K - th order result vector.

[0127] According to an embodiment of the present disclosure, obtaining the k - th order result matrix from the N k - th order feature matrices includes obtaining the k - th order result matrix from the N k - th order feature matrices through a convolutional neural network or a fully - connected network; obtaining the k - th order result vector based on the k - th order result matrix includes performing pooling processing on the k - th order result matrix to obtain the k - th order result vector.

[0128] According to an embodiment of the present disclosure, obtaining the non - linear combination result based on the 0 - th order result vector to the K - th order result vector includes: concatenating the 0 - th order to the K - th order result vectors to obtain a concatenated vector; inputting the concatenated vector into a first fully - connected network to obtain the non - linear combination result. According to an embodiment of the present disclosure, the first fully - connected network includes a fully - connected layer with an activation function.

[0129] As shown in Figure 2As shown, for each of the N vectors, an outer product operation is first performed with all vectors including itself, and N feature matrices are obtained after the operation. The N feature matrices are stacked to form a feature cube. Modules such as CNN or a fully connected network are used to map the feature cube into a new feature matrix as the 0th-order result matrix, and the outer product operation, stacking, and mapping are performed again with the original N vectors to obtain the 1st-order result matrix, and so on. For the kth-order result matrix, the kth-order result vector is obtained by pooling. Then, the 0th-order to Kth-order result vectors are spliced to obtain a spliced vector; the spliced vector is input into the first fully connected network to obtain the nonlinear combination result.

[0130] According to an embodiment of the present disclosure, obtaining the semantic information representation of the conversation text vector through the third model includes: inputting the conversation text vector into a second fully connected network to obtain the semantic information representation of the conversation text vector.

[0131] like Figure 2 As shown, according to an embodiment of the present disclosure, the conversation text vector includes a previous text vector and / or a current text vector. The previous text vector and the current text vector can be concatenated and input into a third model (e.g., a second fully connected network). Alternatively, only the previous text vector or only the current text vector can be input into the second fully connected network, and the second fully connected network outputs a semantic information representation of the conversation text vector. According to an embodiment of the present disclosure, the second fully connected network can be a two-layer fully connected network, wherein the first fully connected layer can have an activation function.

[0132] According to an embodiment of the present disclosure, the fourth model is used to determine the probability that the user performs a behavior related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation, including: inputting the linear combination result, the nonlinear combination result, and the semantic information representation into a third fully connected network to obtain the probability that the user performs a behavior related to the object.

[0133] like Figure 2 As shown, the fourth model can be, for example, a third fully connected network, which includes, for example, a fully connected layer with an activation function. The linear combination result, the nonlinear combination result, and the semantic information representation are input into the third fully connected network to obtain the probability of the user performing an action related to the object.

[0134] According to an embodiment of the present disclosure, the data processing method also includes: obtaining sample data, the sample data including object feature data of the sample object, user portrait data of the sample user, conversation text data of the sample user, and a record of whether the user performs actions related to the object; and using the sample data to train the first model, the second model, the third model, and the fourth model.

[0135] According to an embodiment of the present disclosure, object feature data of a sample object, user profile data of a sample user, conversation text data of the sample user, and a record of whether the sample user has performed any behavior related to the object can be obtained from a database, and data cleaning can be performed, including removing abnormal data, privacy data, etc. Each cleaned sample data includes the object feature data of a sample object, user profile data of a sample user, conversation text data of the sample user, and a record of whether the sample user has performed any behavior related to the object. If the sample user has performed any behavior related to the object, the record can be 1, and the sample data containing the record is positive sample data; if the sample user has not performed any behavior related to the object, the record can be 0, and the sample data containing the record is negative sample data. All positive and negative samples are collected and randomly divided into a training set, a validation set, and a test set. The training set is used to train the first to fourth models, and the validation set and test set are used to evaluate the effectiveness of the trained models.

[0136] Figure 3 A structural block diagram of a data processing device according to an embodiment of the present disclosure is shown, wherein the device can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0137] like Figure 3 As shown, the data processing device 300 includes a first acquisition module 310, a second acquisition module 320, and a determination module 330, wherein:

[0138] The first acquisition module 310 is configured to acquire object feature data of an object, user portrait data of a user, and conversation text data of the user;

[0139] The second acquisition module 320 is configured to acquire an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data;

[0140] The determination module 330 is configured to determine the probability that the user performs an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector.

[0141] According to an embodiment of the present disclosure, the object is a commodity, and the behavior is purchasing or clicking the commodity; or the object is a service, and the behavior is purchasing the service.

[0142] According to an embodiment of the present disclosure, obtaining the object feature vector corresponding to the object feature data includes:

[0143] If the object feature data belongs to category data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data;

[0144] If the object feature data is numerical data, an object feature vector corresponding to the object feature data is determined according to the value of the object feature data and a first preset random vector.

[0145] The object feature vectors corresponding to different object feature data are different.

[0146] According to an embodiment of the present disclosure, obtaining a user portrait vector corresponding to the user portrait data includes:

[0147] If the user portrait data belongs to category data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data;

[0148] If the user portrait data is numerical data, a user portrait vector corresponding to the user portrait data is determined according to the value of the user portrait data and a second preset random vector.

[0149] Among them, the user portrait vectors corresponding to different user portrait data are different.

[0150] According to an embodiment of the present disclosure, obtaining a conversation text vector corresponding to the conversation text data includes:

[0151] Obtaining a current text vector corresponding to a current dialogue text in the dialogue text data;

[0152] A previous text vector corresponding to a previous dialogue text preceding the current dialogue text in the dialogue text data is obtained.

[0153] According to an embodiment of the present disclosure, determining the probability of the user performing an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector includes:

[0154] Through the first model, obtain the first projection vector of the object feature vector in the latent space of the dialogue text vector, the second projection vector of the user portrait vector in the latent space, and the linear combination result of the dialogue text vectors;

[0155] Through the second model, obtain the non-linear combination result of the first projection vector, the second projection vector, and the dialogue text vector;

[0156] Through the third model, obtain the semantic information representation of the dialogue text vector;

[0157] Through the fourth model, determine the probability that the sample user performs an action related to the object according to the linear combination result, the non-linear combination result, and the semantic information representation.

[0158] According to an embodiment of the present disclosure, the step of obtaining the linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the first model includes:

[0159] Through the first model, assign randomly initialized and trainable weight coefficients to each of the first projection vector, the second projection vector, and the dialogue text vector, and output the weighted linear combination result of each vector in the first projection vector, the second projection vector, and the dialogue text vector as the linear combination result.

[0160] According to an embodiment of the present disclosure, the first projection vector, the second projection vector, and the dialogue text vector include N vectors. The step of obtaining the non-linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the second model includes:

[0161] For k = 0, 1, …… K, where K>1: perform an outer product operation on each vector in the k-th order initial matrix with the N vectors respectively to obtain N k-th order feature matrices; obtain the k-th order result matrix from the N k-th order feature matrices, and obtain the k-th order result vector based on the k-th order result matrix; when k<K, use the k-th order result matrix as the (k + 1)-th order initial matrix, where the 0-th order initial matrix includes the N vectors;

[0162] Based on the 0-th order result vector to the K-th order result vector, obtain the non-linear combination result.

[0163] According to an embodiment of the present disclosure, the step of obtaining the k-th order result matrix from the N k-th order feature matrices includes obtaining the k-th order result matrix from the N k-th order feature matrices through a convolutional neural network or a fully connected network;

[0164] The obtaining of the k-th order result vector based on the k-th order result matrix includes performing pooling processing on the k-th order result matrix to obtain the k-th order result vector.

[0165] According to an embodiment of the present disclosure, obtaining the nonlinear combination result based on the 0th order result vector to the Kth order result vector includes:

[0166] Splicing the 0th to Kth order result vectors to obtain a spliced vector;

[0167] The concatenated vector is input into a first fully connected network to obtain the nonlinear combination result.

[0168] According to an embodiment of the present disclosure, obtaining the semantic information representation of the conversation text vector through the third model includes:

[0169] The conversation text vector is input into a second fully connected network to obtain a semantic information representation of the conversation text vector.

[0170] According to an embodiment of the present disclosure, determining, by the fourth model, the probability that the user performs an action related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation includes:

[0171] The linear combination result, the nonlinear combination result, and the semantic information representation are input into a third fully connected network to obtain a probability that the user performs an action related to the object.

[0172] According to an embodiment of the present disclosure, the data processing device 300 further includes:

[0173] The third acquisition module 340 is configured to acquire sample data, wherein the sample data includes object feature data of the sample object, user portrait data of the sample user, conversation text data of the sample user, and a record of whether the user performs any behavior related to the object;

[0174] The training module 350 is configured to train the first model, the second model, the third model, and the fourth model using the sample data.

[0175] The present disclosure also discloses an electronic device, Figure 4 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0176] like Figure 4 As shown, the electronic device 400 includes a memory 401 and a processor 402, wherein the memory 401 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 402 to implement the method according to the embodiment of the present disclosure.

[0177] Specifically, the data processing method includes:

[0178] Obtaining object feature data of an object, user portrait data of a user, and conversation text data of the user;

[0179] Obtaining an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data;

[0180] Based on the object feature vector, the user portrait vector, and the conversation text vector, a probability of the user performing an action related to the object is determined.

[0181] According to an embodiment of the present disclosure, the object is a commodity, and the behavior is purchasing or clicking the commodity; or the object is a service, and the behavior is purchasing the service.

[0182] According to an embodiment of the present disclosure, obtaining the object feature vector corresponding to the object feature data includes:

[0183] If the object feature data belongs to category data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data;

[0184] If the object feature data is numerical data, an object feature vector corresponding to the object feature data is determined according to the value of the object feature data and a first preset random vector.

[0185] The object feature vectors corresponding to different object feature data are different.

[0186] According to an embodiment of the present disclosure, obtaining a user portrait vector corresponding to the user portrait data includes:

[0187] If the user portrait data belongs to category data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data;

[0188] If the user portrait data is numerical data, a user portrait vector corresponding to the user portrait data is determined according to the value of the user portrait data and a second preset random vector.

[0189] Among them, the user portrait vectors corresponding to different user portrait data are different.

[0190] According to an embodiment of the present disclosure, obtaining a conversation text vector corresponding to the conversation text data includes:

[0191] Obtaining a current text vector corresponding to a current dialogue text in the dialogue text data;

[0192] A previous text vector corresponding to a previous dialogue text preceding the current dialogue text in the dialogue text data is obtained.

[0193] According to an embodiment of the present disclosure, determining the probability of the user performing an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector includes:

[0194] Obtaining, through the first model, a linear combination result of a first projection vector of the object feature vector in the latent space of the conversation text vector, a second projection vector of the user portrait vector in the latent space, and the conversation text vector;

[0195] Obtaining a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector through a second model;

[0196] Obtaining semantic information representation of the conversation text vector through a third model;

[0197] The probability that the sample user performs a behavior related to the object is determined through a fourth model according to the linear combination result, the nonlinear combination result, and the semantic information representation.

[0198] According to an embodiment of the present disclosure, obtaining a linear combination result of the first projection vector, the second projection vector, and the conversation text vector through the first model includes:

[0199] Through the first model, a randomly initialized and trainable weight coefficient is assigned to each vector in the first projection vector, the second projection vector, and the conversation text vector, and a weighted linear combination result of the first projection vector, the second projection vector, and each vector in the conversation text vector is output as the linear combination result.

[0200] According to an embodiment of the present disclosure, the first projection vector, the second projection vector, and the conversation text vector include N vectors, and obtaining a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector using the second model includes:

[0201] For k = 0, 1, …… K, where K > 1: perform an outer product operation on each vector in the k-th order initial matrix with each of the N vectors respectively to obtain N k-th order feature matrices; obtain a k-th order result matrix from the N k-th order feature matrices, and obtain a k-th order result vector based on the k-th order result matrix; when k < K, use the k-th order result matrix as the (k + 1)-th order initial matrix, where the 0-th order initial matrix includes the N vectors.

[0202] Obtain the non-linear combination result based on the 0-th order result vector to the K-th order result vector.

[0203] According to an embodiment of the present disclosure, obtaining the k-th order result matrix from the N k-th order feature matrices includes obtaining the k-th order result matrix from the N k-th order feature matrices through a convolutional neural network or a fully connected network.

[0204] Obtaining the k-th order result vector based on the k-th order result matrix includes performing pooling processing on the k-th order result matrix to obtain the k-th order result vector.

[0205] According to an embodiment of the present disclosure, obtaining the non-linear combination result based on the 0-th order result vector to the K-th order result vector includes:

[0206] Concatenate the 0-th order to the K-th order result vectors to obtain a concatenated vector.

[0207] Input the concatenated vector into a first fully connected network to obtain the non-linear combination result.

[0208] According to an embodiment of the present disclosure, obtaining the semantic information representation of the dialogue text vector through a third model includes:

[0209] Input the dialogue text vector into a second fully connected network to obtain the semantic information representation of the dialogue text vector.

[0210] According to an embodiment of the present disclosure, determining the probability that the user makes an action related to the object through a fourth model based on the linear combination result, the non-linear combination result, and the semantic information representation includes:

[0211] Input the linear combination result, the non-linear combination result, and the semantic information representation into a third fully connected network to obtain the probability that the user makes an action related to the object.

[0212] According to an embodiment of the present disclosure, the method further includes:

[0213] Acquire sample data, the sample data including object feature data of a sample object, user portrait data of a sample user, conversation text data of the sample user, and a record of whether the user performs any behavior related to the object;

[0214] The first model, the second model, the third model, and the fourth model are trained using the sample data.

[0215] Figure 5 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown.

[0216] like Figure 5 As shown, the computer system 500 includes a processing unit 501, which can execute various processes in the above-mentioned embodiments according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage unit 508 into a random access memory (RAM) 503. Various programs and data required for the operation of the system 500 are also stored in the RAM 503. The processing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0217] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed so that a computer program read therefrom can be installed into the storage section 508 as needed. Among them, the processing unit 501 can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.

[0218] In particular, according to embodiments of the present disclosure, the methods described above can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a computer-readable medium, the computer program comprising program code for executing the methods described above. In such embodiments, the computer program can be downloaded and installed from a network via the communication portion 509 and / or installed from the removable medium 511.

[0219] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and the module, program segment or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0220] The units or modules involved in the embodiments described in this disclosure may be implemented by software or programmable hardware. The units or modules described may also be provided in a processor, and the names of these units or modules do not, in certain circumstances, constitute limitations on the units or modules themselves.

[0221] As another aspect, the present disclosure further provides a computer-readable storage medium. This computer-readable storage medium may be included in the electronic device or computer system described in the above embodiments, or may be a standalone computer-readable storage medium not incorporated into the device. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the methods described in the present disclosure.

[0222] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention herein is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims

1. A data processing method, comprising: Obtaining object feature data of an object, user portrait data of a user, and conversation text data of the user; Obtaining an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data; Determining, based on the object feature vector, the user portrait vector, and the conversation text vector, a probability that the user performs an action related to the object; Wherein, the method of determining the probability that the user performs a behavior related to the object based on the object feature vector, the user portrait vector, and the conversation text vector includes: obtaining, through a first model, a linear combination result of a first projection vector of the object feature vector in the latent space of the conversation text vector, a second projection vector of the user portrait vector in the latent space, and the conversation text vector; obtaining, through a second model, a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector; obtaining, through a third model, a semantic information representation of the conversation text vector; and determining, through a fourth model, the probability that the user performs a behavior related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation.

2. The method according to claim 1, wherein The acquiring of the object feature vector corresponding to the object feature data includes: If the object feature data belongs to category data, mapping the object feature data to a corresponding random vector according to the value of the object feature data, and using the corresponding random vector as the object feature vector corresponding to the object feature data; If the object feature data is numerical data, an object feature vector corresponding to the object feature data is determined according to the value of the object feature data and a first preset random vector. The object feature vectors corresponding to different object feature data are different.

3. The method according to claim 1, wherein The obtaining of a user portrait vector corresponding to the user portrait data includes: If the user portrait data belongs to category data, mapping the user portrait data to a corresponding random vector according to the value of the user portrait data, and using the corresponding random vector as the user portrait vector corresponding to the user portrait data; If the user portrait data is numerical data, a user portrait vector corresponding to the user portrait data is determined according to the value of the user portrait data and a second preset random vector. Among them, the user portrait vectors corresponding to different user portrait data are different.

4. The method according to claim 1, wherein The obtaining of a conversation text vector corresponding to the conversation text data includes: Obtaining a current text vector corresponding to a current dialogue text in the dialogue text data; A previous text vector corresponding to a previous dialogue text preceding the current dialogue text in the dialogue text data is obtained.

5. The method according to claim 1, wherein The obtaining, through the first model, a linear combination result of the first projection vector, the second projection vector, and the conversation text vector includes: Through the first model, assign randomly initialized and trainable weight coefficients to each of the first projection vector, the second projection vector, and the dialogue text vector, and output the weighted linear combination result of each vector in the first projection vector, the second projection vector, and the dialogue text vector as the linear combination result.

6. The method according to claim 1, wherein The first projection vector, the second projection vector, and the dialogue text vector include N vectors. The method for obtaining the non-linear combination result of the first projection vector, the second projection vector, and the dialogue text vector through the second model includes: For k = 0, 1, …… K, where K>1, perform an outer product operation on each vector in the k-th order initial matrix with the N vectors respectively to obtain N k-th order feature matrices; obtain the k-th order result matrix from the N k-th order feature matrices, and obtain the k-th order result vector based on the k-th order result matrix; when k < K, use the k-th order result matrix as the (k + 1)-th order initial matrix, where the 0-th order initial matrix includes the N vectors; Based on the 0-th order result vector to the K-th order result vector, obtain the non-linear combination result.

7. The method according to claim 6, wherein: The obtaining the k-th order result matrix from the N k-th order feature matrices includes obtaining the k-th order result matrix from the N k-th order feature matrices through a convolutional neural network or a fully connected network; [[ID=]]The obtaining the k-th order result vector based on the k-th order result matrix includes performing pooling processing on the k-th order result matrix to obtain the k-th order result vector.

8. The method according to claim 6, wherein: The obtaining the non-linear combination result based on the 0-th order result vector to the K-th order result vector includes: Concatenate the 0-th order to the K-th order result vectors to obtain a concatenated vector; Input the concatenated vector into the first fully connected network to obtain the non-linear combination result.

9. The method according to claim 1, wherein The obtaining the semantic information representation of the dialogue text vector through the third model includes: Input the dialogue text vector into the second fully connected network to obtain the semantic information representation of the dialogue text vector.

10. The method according to claim 1, wherein The determining the probability that the user makes a behavior related to the object according to the linear combination result, the non-linear combination result, and the semantic information representation through the fourth model includes: Input the linear combination result, the non-linear combination result, and the semantic information representation into the third fully connected network to obtain the probability that the user makes a behavior related to the object.

11. The method according to claim 1, further comprising: Obtain sample data, where the sample data includes the object feature data of the sample object, the user profile data of the sample user, the dialogue text data of the sample user, and the record of whether the user makes a behavior related to the object; Use the sample data to train the first model, the second model, the third model, and the fourth model.

12. The method according to claim 1, wherein: The object is a commodity, and the behavior is to purchase or click on the commodity; or The object is a service, and the behavior is to purchase the service.

13. A data processing device comprising: A first acquisition module is configured to acquire object feature data of an object, user portrait data of a user, and conversation text data of the user; A second acquisition module is configured to acquire an object feature vector corresponding to the object feature data, a user portrait vector corresponding to the user portrait data, and a conversation text vector corresponding to the conversation text data; a determination module configured to determine a probability that the user performs an action related to the object based on the object feature vector, the user portrait vector, and the conversation text vector; Wherein, the determination module is further configured to obtain, through a first model, a linear combination result of the first projection vector of the object feature vector in the latent space of the conversation text vector, the second projection vector of the user portrait vector in the latent space, and the conversation text vector; obtain, through a second model, a nonlinear combination result of the first projection vector, the second projection vector, and the conversation text vector; obtain, through a third model, a semantic information representation of the conversation text vector; and determine, through a fourth model, the probability that the user will perform an action related to the object based on the linear combination result, the nonlinear combination result, and the semantic information representation.

14. The apparatus according to claim 13, further comprising: a fourth acquisition module configured to acquire sample data, the sample data including object feature data of a sample object, user portrait data of a sample user, conversation text data of the sample user, and a record of whether the user performs an action related to the object; A training module is configured to train the first model, the second model, the third model, and the fourth model using the sample data.

15. An electronic device comprising a memory and a processor; wherein: The memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method steps according to any one of claims 1 to 12.

16. A readable storage medium having computer instructions stored thereon, wherein the computer instructions, when executed by a processor, implement the method steps according to any one of claims 1 to 12.

Images